A recommendation can be a useful lead while still lacking evidence for one of your requirements. Treat missing information as unresolved, especially when the condition is essential. A matching name or topic does not verify every attribute around it.
OnlyDetect's September 30, 2026 directory snapshot makes that distinction concrete: the eligible sample had 3,625 tag assignments, but none carried the verification marker counted in the audit.
What the snapshot actually measures
The published data file reports 461 eligible directory profiles. Of those, 447 had at least one tag assignment. The audit counted zero verified tag assignments.
This is a convenience sample of active, indexable, unsuppressed OnlyDetect records validated within the study's stated window. It is not a representative sample of all OnlyFans accounts. A missing verification marker also does not prove a tag false.
It means a stored label and a verified label are different states. See the filter-coverage study for the audit's scope and interpretation.
Why this matters for recommendations
In the implementation reviewed on September 30, OnlyDetect's recommendation candidate query includes only tags with a recorded verification timestamp in its returned tag list. Other public fields, such as a biography, can still contribute information.
That is a rule found in code, not a controlled model-accuracy experiment. It does not establish that every recommendation satisfies every request or that all visible directory tags have been independently checked.
The methodology explains the underlying public-source boundary and the importance of provenance.
Sort your requirements into three states
For each candidate, use a small decision table:
Scroll across the table on smaller screens.
| Requirement | Evidence state | Action |
|---|---|---|
| Publicly stated interest | Relevant current public wording found | Keep the source with the note |
| Preferred broad region | Missing or ambiguous | Leave unresolved |
| Essential offer condition | Not supported by the current offer | Do not treat the candidate as confirmed suitable |
This is an editorial evaluation method. It is not a promise that an AI interface exposes all three states automatically.
Suppose a fictional request asks for photography, a paid price below $15, and fast personal replies. A biography may support photography. A recorded price may help shortlist the account. Neither establishes fast personal replies.
Decide whether unknown is acceptable
A preference can remain unknown while you continue exploring. An essential condition needs supporting evidence before you rely on it. Define that distinction yourself rather than letting a fluent recommendation decide it for you.
If the AI discovery tool returns a candidate, open the actual profile and verify the conditions that matter. Do not interpret the absence of a warning as proof that every condition passed.
The general AI verification guide explains claim-by-claim checks. This case adds a specific lesson from real directory data: the presence of a field is not the same as evidence that the field was verified.